Description
I encountered a TypeError when running the rollout with SGLangHttpServer. It seems that _launch_subprocesses() is being called with missing required arguments.
Error Message
TypeError: _launch_subprocesses() missing 3 required positional arguments:
'init_tokenizer_manager_func',
'run_scheduler_process_func',
'run_detokenizer_process_func'
Stack Trace (excerpt)
File ".../async_sglang_server.py", line 203, in launch_server
self.tokenizer_manager, self.template_manager, self.scheduler_info, *_ = _launch_subprocesses(...)
Environment
- Python: 3.10
- Ray: 2.54.1
- verl: 0.7.0.dev0
- sglang: 0.5.9
Additional Context
It looks like there may be a mismatch between the expected signature of _launch_subprocesses() and how it is being called in async_sglang_server.py. Possibly caused by a version inconsistency between modules.
Question
Could I please ask if there is a specific version combination required for verl and sglang?
Thanks for your help!
Description
I encountered a
TypeErrorwhen running the rollout withSGLangHttpServer. It seems that_launch_subprocesses()is being called with missing required arguments.Error Message
Stack Trace (excerpt)
Environment
Additional Context
It looks like there may be a mismatch between the expected signature of
_launch_subprocesses()and how it is being called inasync_sglang_server.py. Possibly caused by a version inconsistency between modules.Question
Could I please ask if there is a specific version combination required for
verlandsglang?Thanks for your help!